Date: Tue, 27 Apr 2010 14:34:22 +0300 From: Andriy Gapon <avg@icyb.net.ua> To: Jeff Roberson <jroberson@jroberson.net> Cc: Bruce Cran <bruce@cran.org.uk>, Alex Keda <admin@lissyara.su>, freebsd-current@freebsd.org Subject: Re: HEADS UP: SUJ Going in to head today Message-ID: <4BD6CBBE.3030901@icyb.net.ua> In-Reply-To: <alpine.BSF.2.00.1004261959480.1398@desktop> References: <4BD35437.2060208@lissyara.su> <alpine.BSF.2.00.1004241656270.1398@desktop> <622DDEDF-0320-49DA-8037-CA8C1F682CC1@samsco.org> <201004251955.03492.bruce@cran.org.uk> <alpine.BSF.2.00.1004261959480.1398@desktop>
next in thread | previous in thread | raw e-mail | index | archive | help
on 27/04/2010 09:00 Jeff Roberson said the following:
> I think some people are enabling after returning to single user from a
> live system rather than booting into single user. This is a different
> path in the filesystem as booting directly just mounts read-only while
> the other option updates a mount from read/write. I believe this is the
> path that is broken.
Yes, this seems to be broken and perhaps by design.
g_vfs_open() calls g_access like this: g_access(cp, 1, wr, 1);
That means that 'e' count (exclusive) is always bumped, even for R/O mounts, and
that prevents opening the provider for writing.
ffs_mountfs has this special code:
/*
* If we are a root mount, drop the E flag so fsck can do its magic.
* We will pick it up again when we remount R/W.
*/
if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS))
error = g_access(cp, 0, 0, -1);
So, basically for read-only UFS root mount we allow concurrent open, even for
writing. This is needed primarily for fsck, but also helps tunefs.
But I believe that this code is exercised only during original mount.
Remounting to R/O at later time doesn't drop 'e' count.
I think that this is by design, to prevent foot-shooting.
We either should document this behavior or re-consider it.
--
Andriy Gapon
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4BD6CBBE.3030901>